Dear all, 

I am setting up a "reversed" VSLM- analysis  so I have a binarized inverted risk factor for stroke and binary lesion maps. When I enter this paradigm to NPM it runs fine and brings up reasonable results in FDR-corrected threshold maps. In Niistat the SAME analysis leads to the output "q=0.05 FDR correction for variable with 130000 tests is z <-Inf, z> Inf". This results in 0 voxels surviving for each bin of the variable or does "Inf" actually mean that threshold is above the z-level of my maps?. May the issue therefor be, that the threshold for FDR- calculation crashes? Do you think it is still appropriate to use NPM or are there big discrepancies in the basic statistic implementation or statistical Bugfixes. 

Best regards Matthias